I want to show that, when $\mu=\mu_0$, then gamma family $\Gamma(a,b)$ is a conjugate prior to inverse Gaussian with density $f(x,\mu,\lambda)=\sqrt{\frac{\lambda}{2\pi x^2}}exp[-\frac{\lambda(x-\mu)^2}{2\mu^2x}]$.

I was able to verify that $IG$ is an exponential family generated by $\bf{T}$$(X)=-1/2(X,X^{-1})^T$ and $h(x)=\sqrt{\frac{1}{2\pi x^3}}$ but I am pretty confused on how to show the conjugate prior.

Any helps would be deeply appreciated!!

  • 1
    $\begingroup$ The conjugate prior for an exponential family $f(x;\theta)=h(x)\exp\{\theta T(x)-\psi(\theta)\}$ is $\pi(\lambda;a,b)\propto\exp\{\theta a-b\psi(\theta)\}$, the measure $h(\cdot)$ does not matter. $\endgroup$ – Xi'an Mar 9 at 7:10
  • $\begingroup$ When $\mu=\mu_0$, a fixed value, the sufficient statistic is no longer $(X,X^{-1})$. $\endgroup$ – Xi'an Mar 9 at 7:12
  • $\begingroup$ @Xi'an Yeah, I know the formula. I have trouble deriving it. $\endgroup$ – statwoman Mar 9 at 14:54

$$\pi_t(\theta)=exp(\Sigma_{j=1}^k\eta_j(\boldsymbol{\theta})t_j-t_{k+1}B(\boldsymbol{\theta})-logw(\mathbf{t}))$$ $$w(t)=\int...\int\int_Rexp(\Sigma t_j\eta_j-t_{k+1}B)d\theta_1...d\theta_k$$ $$B(\theta)=-\frac{\lambda}{\mu}-\frac{log\lambda}{2}$$ Since $\mu$ is known then k=1 in this part. taking $a=t1$ and $b=\frac{t2-t1}{\mu}$, we yield: $$w(t)=\int_0^\infty exp(-b\lambda)\lambda^{a/2}d\lambda=b^{-\frac{a+1}{2}}\Gamma(\frac{a+1}{2})$$ $$\Pi_t(\theta)=\frac{\lambda^{\frac{a+1}{2}}e^{-b\lambda}}{\Gamma(\frac{a+1}{2})}b^{\frac{a+1}{2}}$$ We can conclude that this shows the gamma distribution with parameters $\alpha=\frac{a+1}{2}$ and $\beta=b$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.